Image Mosaic Techniques for the Restoration of Virtual Heritage 2003. 8. 28 Yong-Moo Kwon, Ig-Jae Kim, Tae-Sung Lee, Se-Un Ryu, Jae-Kyung Seol KIST KOREA Contents Revisiting Image Mosaic Technique Our Researches for Image Mosaic IR Reflectography Image Mosaic X-Ray Image Mosaic Summary Revisiting Image Mosaic Technique Image Mosaicing Panorama Image Image Based Rendering (IBR) Basic Algorithm Registration using Features Image Warping based on Homography Matrix Blending Images Target Dimension in view of Image Mosaicing 2D Target Planar Paintings Image Homography Technique Feature-Based Image Mosaicing 3D Target 3D Real World Image Limitation using Homography Due to Depth Difference b/w Features in Target Our Research for Image Mosaic 2D Target IR Reflectography Special 3D Target X-Ray Imaging Mural Underdrawings Mosaic Old Sword X-Ray Image Mosaic Research Topics How to extract and use Features Imaging Media (IR, X-Ray) Feature’s characteristics are different from the previous ones IR Image Mosaic IR Reflectography System IR Source IR Filter IR Camera Murals IR Reflectography Principle Visible Light Reflection UnderDrawing IR Color Painting, Reflection Dust Absorbed Back Frame IR Reflectography Camera IR Camera : Super eye C2847 (~1.9㎛) Hamamatsu IR Source : ~1.9㎛ IR Filter : CVI Laser Corp. NIR bandwidth filter Bandpass Filter 800nm ~ 2000nm Pass every 100 nm bandpass filter (800nm, 900, …, 2000nm) IR Characteristics to Mural according to WL IR Camera HAMAMATSU Super eye C2847 WL Range : 0.4㎛ ~ 1.9㎛ IR Source HAMAMATSU C1385-02 Filter HAMAMATSU IR-D80A : 0.8㎛ ~ 1.9㎛ CVI Laser corporation Near IR Interference BP filter 800nm, 900nm, … 2000nm Sony PC-115 Digital Image Capture Night Shot IR Image Mosaic for Mural Underdrawing Basic Method Automatic Feature Extraction Registration Using Features Image Warping Image Blending Main Considerations IR Wavelength Characteristics Penetration Ratio into Paintings Color (Red, Green, Blue etc) Color Painting Depth Our Approach ▶ Automatic Feature Extraction & Registration - Cross Points in IR Underdrawing Image - Grid Pattern for Blank Space ▶ Adaptive Overlapping Area For Image Blending - Trade-Off between Registration and Blending * Large Overlapping Area: Good for Registration * Small Overlapping Area: Good for Blending ▶ Use feature of IR Spectrum - Use Different IR Wavelength according to paining color Automatic Feature Extraction Feature of Korea Murals - Many Blank Space - Not so much good features 1> Visible Light Pattern 2> Twice Captures - w/o IR Filter - w/i IR Filter IR Image Mosaic - Homography Estimation using Grid Image & IR Image - Apply Homography to IR Image X-RAY Image Mosaic Why we use X-ray Technique ? Old Sword Old Sword is inside Sword Cover Weak for Touch & Manipulation Can’t Open Sword Cover Use X-Ray Technique for the restoration of Old Sword inside Sword Cover Sword for experiment Schema of a x-ray imaging using a linear X-Ray Camera 1. X-Ray Image 2. X-Ray Tube 3. X-Rays 4. X-Ray Detector 5. PC 6. Object Why X-Ray Image Mosaic ? For High Resolution Imaging Multiple X-Ray Imaging Setting Object X-Ray Image Capture Move Object Upward or Downward Step-By-Step Stitching X-Ray Images into High Resolution Image X-Ray Imaging Principle Basic Principle X-Ray Particle Penetrates through Target One Point Depth -> Grey Value Pixel Dependency Target Depth Target Material X-Ray Image Characteristics: 2D or 3D ? Target Dimension in view of Image Mosaic Well Controlled Penetration Angle Image Pixel Depends on Penetration Angle Usually Same Penetration Angle for Each Capture Just Planar 2D Image Using CCD Camera Orthogonal axis Movement according to X-Ray Beam Object -> X-Ray Camera -> CCD Camera 2D Target: Homography Technique X-RAY Image Equipment X-TEK X-Ray System X-Ray Source & Object (Sword) X-RAY Image Capture For High-Resolution Restoration Multiple X-Ray Imaging Image Stitching Technique Feature-based Registration Problem ? Difficult to use features in X-ray Image Using Feature Pattern Feature Extraction Feature Extraction From Known Pattern Circle Type & Rectangular Type Circle Type -> Pattern Matching Rectangular -> Feature Points Feature Ex traction Binary pattern for feature ID Feature Extraction Method Circle Type Pattern -> Apply Image Labeling Rectangular Type Pattern -> Corner Detection D x2 C D x D y D x D y 2 - For every pixel of image, computes first derivatives Dx and Dy. - The eigenvalues are found by solving det(C- λI )= 0 DxDy 2 D y 2 ( D x D y ) 4( D x 2 2 2 2 2 If λ1, λ2 > t, where t is some threshold, then a corner is found at that location D y ( D x D y ) ) 2 2 Feature Point Matching Semi-Auto(Present) Automatic Matching (On-going) Automatic Feature Extraction of Rectangle Type pattern Manual Matching Classify the features using pattern ID from Circle Type Pattern Homography Matrix Apply LS-Method(Least Square Method) using Matched feature Points Semi-auto Demo Implemented S/W X-ray Image File Handling Feature Extraction & Select Points Homograph y Matrix Estimation & Stitching Generated HighResolution X-ray Image More Experimentation Summary Application of Image Mosaicing Techniques Infrared Image X-Ray Image Our Approach Feature Pattern Automatic Feature Extraction & Registration Homography Technique Imaging Media (IR, X-Ray) & Feature’s Characteristics Thank You ! ymk@cherry.kist.re.kr